A Weighted Distance Approach to Relevance Feedback

Content-based image retrievalsystems use low-levelfeatures like color and texturefor image representation. Given these representationsasfeature vectors, similarity between images is measured by computing distances in thefeature space. Unfortunately, these low-levelfeatures cannot always capture the high-level concept of similarity in human perception. Relevancefeedback tries to improve theperformance by allowing iterative retrievals where thefeedback informationfrom the user is incorporated into the database search. We present a weighted distance approach where the weights are the ratios of standard deviations of thefeature values both for the whole database and also among the images selected as relevant by the user. Thefeedback is used for both independent and incremental updating of the weights and these weights are used to iteratively refine the effects of differentfeatures in the database search. Retrieval performance is evaluated using average precision and progress that are comp...

• Type: Conference
• Year: 2000
• Where: ICPR
• Authors: Selim Aksoy, Robert M. Haralick, Faouzi Alaya Cheikh, Moncef Gabbouj
• Keywords: Average Performance Improvement, Computer Vision, Database Search, ICPR 2000, Weighted Distance Approach